Finding the Real
Problem Before
Writing a Line
of Code

Clayton Christensen spent decades trying to get companies to ask a different question. Not "what does our customer want?" but "what job is our customer hiring this product to do?" The distinction sounds academic until you build something for the wrong job and wonder why nobody showed up. This case study is about finding the right job before the building started.

The Market, Briefly

Nigeria has somewhere in the region of 40 million small and medium-sized businesses. The overwhelming majority handle payroll the same way: a spreadsheet, a list of bank account numbers, and a few hours of manual transfers at the end of every month. No dedicated software. No HR function. Just the owner or someone trusted, doing it by hand.

The opportunity, on the surface, looks obvious. Manual payroll is slow and error-prone. Automate it. That was the hypothesis going in — and it turned out to be directionally correct about the problem, and significantly wrong about what made the problem urgent. Getting that distinction right early, before any product decisions were made, was the entire point of what followed.

Why the Assumption Needed Testing

Steve Blank has a line that practitioners in early-stage ventures quote often enough that it risks losing its edge: "No business plan survives first contact with customers." What he means, more precisely, is that the plan is built on assumptions, and assumptions are not evidence. They are starting points for investigation.

The starting assumption here was that Nigerian SME owners were primarily suffering from operational friction. Payroll takes too long. Manual transfers are tedious. Errors slip through. The solution would be a self-serve platform that compressed the process — calculation, payment, record-keeping — into something manageable without specialist support.

That assumption had intuitive force. Manual anything is painful. But intuitive force is not signal. Eric Ries put it plainly: "The only way to win is to learn faster than anyone else." The question was what learning structure would produce the clearest answer the fastest.

How I Got to People

The first round used LinkedIn outreach targeting SME founders, managing directors, and business owners in Lagos, Abuja, and Port Harcourt. Direct messages, peer framing, no survey language.

The design of the conversation mattered more than the channel. There is a well-documented gap between what people say they experience and what actually drives their behaviour — Kahneman called it the difference between the experiencing self and the narrating self. Ask someone about their pain and they give you their story. State the experience and ask if it matches, and they react. The reaction is more honest.

So the opening wasn't "tell me about your payroll process." It was closer to: "Most business owners running payroll manually in Nigeria describe a specific kind of anxiety around PAYE deductions — not just the manual work, but the uncertainty about whether they've done it correctly. Does that match your experience?" If yes: walk me through the last time that felt real. If no: where does the actual friction show up for you?

Every follow-up was determined by what the person said, not by a script. The goal was not to build a picture of the market. It was to determine whether real people could recall real incidents where the problem had cost them something — time, money, confidence, or sleep.

Three pain clusters identified in Phase 1 outreach — operational friction, sign-off anxiety, compliance exposure — shown as relative frequency of unprompted mentions across conversations.

What the Conversations Revealed

Three pain clusters surfaced. The first was operational friction — payroll taking too long, transfers being tedious. The second was sign-off anxiety — discomfort approving calculations without certainty. The third was compliance exposure — specific fear around PAYE errors, FIRS and LIRS enforcement, and personal liability for getting statutory remittances wrong.

The compliance cluster was the only one that appeared unprompted and with the kind of specificity that distinguishes real fear from background noise. People described actual scenarios. A tax audit a peer had been through. A penalty letter someone in their network received. A calculation they had second-guessed for weeks before finally submitting.

Nassim Taleb draws a sharp line between noise and signal. Most of what gets called market insight is noise dressed up as pattern. The test for signal is repetition across independent sources without prompting. The compliance fear passed that test. The operational inconvenience did not.

The operational pain was real. It was not what was keeping anyone up at night.

The Problem With Stopping There

Qualitative signal from conversations is useful. It is not sufficient. Sample sizes are small. Selection effects are real — the people who respond to LinkedIn outreach are not a random draw from the population. And there is a persistent human tendency, noted by Cialdini among others, to tell researchers what sounds reasonable rather than what is actually driving behaviour.

The conversations pointed clearly toward compliance as the real entry point. But "pointed clearly" is not the same as "confirmed." Two framings existed — compliance risk and operational convenience — and both needed to be tested against the same audience simultaneously, with behaviour as the only judge. Behaviour, not stated preference. Not survey response. Not a conversation.

Testing Both Framings Head-to-Head

A five-day Meta campaign. Two ad creatives, running in parallel. Identical targeting. Identical budget allocation. Identical placement — Facebook and Instagram feeds only, no Audience Network, no Reels. Total spend: $98.13.

The compliance creative led with: "PAYE errors are no longer a small business problem. They are a penalty waiting to happen." The operational pain creative led with: "Every month-end, payroll takes longer than it should. That changes now."

Both pointed to the same landing page. One form field. Email address. Submit. The conversion rate on that page was the primary decision metric — not clicks, not impressions, not engagement. Email submission under low friction represents a voluntary decision to stay connected to something. That is the behaviour that counts.

Ad campaign structure — two creatives × two ad sets (interest targeting vs job title targeting), identical spend and placement per cell.

What the Data Said

The result was not close.

The compliance creative generated 61 of 63 total leads. The operational pain creative received meaningful impressions in both ad sets and produced 2. A 30-to-1 ratio. That is not a performance gap. That is a directional finding. The two framings were not competing — one was speaking to something active and specific, the other was describing a background inconvenience that people tolerate without urgency.

Overall conversion rate across the campaign was 8% — 63 leads from 788 link clicks. Cost per lead: $1.56. The interest-based ad set outperformed the job title ad set: 37 leads at $1.32 CPL versus 26 leads at $1.89 CPL on near-identical spend.

A campaign this small is not statistically decisive in isolation. But Geoffrey Moore made a point about early markets that applies here: the goal in the early stages is not to prove a thesis beyond doubt, it is to reduce uncertainty enough to make a directional commitment. The compliance signal had now appeared, independently and consistently, across two methods: qualitative conversation and paid behavioural testing. That convergence is what you act on.

What the Submitters Confirmed

After submitting their email, respondents received a short follow-up asking about their current payroll situation. Sequencing matters here. People who have already acted are more candid about why. People asked to explain their interest before acting give you what sounds reasonable. The follow-up came after the action.

That last number is the one worth sitting with. Clicking an ad and entering an email costs almost nothing. Staying on a waitlist after being told to wait costs a decision — a small one, but a real one. 86% retention after a deliberate friction point is not casual interest. It is a specific, active problem that people are waiting to have solved.

Philip Kotler spent a career making a version of this point: customers do not buy products, they buy solutions to problems. The question is always how active and specific the problem is. Passive problems generate polite interest. Active problems generate waitlists.

What Was Built

The validation produced a clear enough signal to proceed. The product is Emitroll — a self-serve payroll platform for Nigerian businesses with 1 to 50 employees. PAYE calculation, statutory deductions, employee payments, without requiring an accountant or HR team. The compliance entry point — the thing the validation identified as the real driver — anchors the positioning.

The MVP is in active development.

What This Was Actually Doing

This was not market research in the conventional sense. Paul Graham has a useful way of separating real work from work that feels productive: the same logic applies to validation. Running conversations and launching ads does not in itself produce clarity. The design of those activities — what question each one was answering, what signal would constitute a yes and what would constitute a no — is where the work lives.

Every conversation in Phase 1 was designed to answer one question: is this problem real and specific enough that people can recall it in their own lives? Every element of the ad campaign was designed to answer one question: when exposed to both framings under identical conditions, which one moves people to act?

The build direction came from the answers. The compliance framing was not chosen because it sounded better or felt more compelling. It was chosen because, across two different methods and two separate moments in the process, it was the only framing that produced consistent, specific, unprompted signal from real people describing real situations.